1164259 - We shouldn't use $http for getting resultset information

Status: RESOLVED DUPLICATE of bug 1409103

(Tree Management :: Treeherder: Frontend, defect, P3)

Description

[William Lachance (:wlach)]

Just clearing my patch backlog.

Anyway, like the title says, we shouldn't do this. Even with asynchronous $http apply enabled, doing this results in us doing a bunch of spurious ingest cycles where nothing actually changes. If we're fetching 50 jobs, the overhead of this is substantial (4213ms in my tests on my Linux desktop).

Comment 1

[William Lachance (:wlach)]

Attachment: PR

This speeds things up a bunch (~4000ms on my machine) by avoiding a whack of useless digests when fetching lots of result sets. See PR for more details.

Comment 2

[Mauro Doglio [:mdoglio]]

I think this patch will break places like the similar jobs panel where we expect a digest cycle to run after the http request.
Also, some tests may stop working because the http mock service is integrated in $http afaik.
We should try to experiment with $scope.evalAsync() to run only one digest cycle when all the job requests complete (instead of 10).

Comment 3

[William Lachance (:wlach)]

We discussed this a bit on irc and I think determined that we should do evalAsync in resultsets_store after everything is processed.

However, having refactored things to do this I'm not seeing any performance improvements anymore. Maybe I need to look into this again or maybe this just isn't worth doing. Attached my wip to the PR if anyone wants to take a look, I am personally considering giving up on this approach...

Comment 4

[William Lachance (:wlach)]

I'm not working on this. Not clear if it's worth it.

Comment 5

[Ed Morley [:emorley]]

Usages of $http will go away as part of bug 1409103.